在黄金期货价格预测问题的研究中,价格具有时变性、非线性、高噪声和影响因子复杂等因素,决定了其被准确预测的难度。传统方法对黄金期货价格的预测主要借助于静态模型,导致预测精度不高或分析不足。为了能动态而准确的预测黄金期货价格,本文从技术行情指标、行业方面的影响因素及宏观经济环境指标三个维度选取39个变量,以机器学习(machine learning;ML)方法构建基本融合素材,利用动态模型平均(dynamic model averaging,DMA)方法代替传统模型融合技巧,得到黄金期货价格预测模型。实证结果表明,采用机器学习-动态模型平均策略能够明显提高黄金期货价格的预测精度。
为了充分挖掘影响金融数据中时序性特征联系,提高预测黄金期货价格的预测精度,提出了一种基于粒子群优化(PSO)、时间卷积神经网络(TCN)和双向门控循环单元(BiGRU)模型相结合的优化双智能体深度学习模型的黄金期货价格预测方法。选择2018~2024年交易日成交价,并添加六大类影响因素数据。通过贝叶斯优化XGBoost并结合SHAP算法进行可解释性特征筛选,再建立PSO-TCN-BiGRU混合模型进行预测,并通过与CNN、BiGRU、TCN、BiGRU-TCN、CNN-BiGRU、TCN-BiGRU模型对比分析,结果表明所提预测模型的RMSE值、MSE值、MAE值都低于其他模型,R方值最高,能更准确地预测黄金期货价格走势。The proposed optimized dual-agent deep learning model, based on particle swarm optimization (PSO), time convolutional neural network (TCN), and bidirectional gated recurrent unit (BiGRU) model, aims to fully explore the temporal characteristic relationships affecting financial data and enhance the prediction accuracy of gold futures price. The trading prices of 2018~2024 trading days were selected, along with six categories of influencing factor data. Bayesian optimization XGBoost combined with SHAP algorithm was used to screen interpretability features, followed by the establishment of a PSO-TCN-BiGRU mixed model for prediction. Comparison and analysis with other models such as CNN, BiGRU, TCN, BiGRU-TCN, CNN-BiGRU, TCN-BiGRU models revealed that the proposed forecasting model exhibited lower RMSE value, MSE value and MAE value than other models while achieving the highest R-square value. This indicates its ability to predict the trend of gold futures price more accurately.
